generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added async version of postfilter_textureness and prefilter_xsobel, modified async test and added test for async version of stereobm

Browse Source
This commit is contained in:
Andrey Morozov
2010-09-08 15:13:23 +00:00
parent 31dbefc865
commit 1feb5b4d02
4 changed files with 126 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
modules/gpu/src/cuda/stereobm.cu
View File

@@ -410,7 +410,7 @@ extern "C" __global__ void prefilter_kernel(unsigned char *output, size_t step,
namespace cv { namespace gpu { namespace bm
{
extern "C" void prefilter_xsobel(const DevMem2D& input, const DevMem2D& output, int prefilterCap)
extern "C" void prefilter_xsobel(const DevMem2D& input, const DevMem2D& output, int prefilterCap, const cudaStream_t & stream)
{
cudaChannelFormatDesc desc = cudaCreateChannelDesc<unsigned char>();
cudaSafeCall( cudaBindTexture2D( 0, stereobm_gpu::texForSobel, input.ptr, desc, input.cols, input.rows, input.step ) );
@@ -421,10 +421,18 @@ namespace cv { namespace gpu { namespace bm
grid.x = divUp(input.cols, threads.x);
grid.y = divUp(input.rows, threads.y);
stereobm_gpu::prefilter_kernel<<<grid, threads>>>(output.ptr, output.step, output.cols, output.rows, prefilterCap);
cudaSafeCall( cudaThreadSynchronize() );
if (stream == 0)
{
stereobm_gpu::prefilter_kernel<<<grid, threads>>>(output.ptr, output.step, output.cols, output.rows, prefilterCap);
cudaSafeCall( cudaThreadSynchronize() );
}
else
{
stereobm_gpu::prefilter_kernel<<<grid, threads, 0, stream>>>(output.ptr, output.step, output.cols, output.rows, prefilterCap);
}
cudaSafeCall( cudaUnbindTexture (stereobm_gpu::texForSobel ) );
}
}}}
@@ -532,7 +540,7 @@ extern "C" __global__ void textureness_kernel(unsigned char *disp, size_t disp_s
namespace cv { namespace gpu { namespace bm
{
extern "C" void postfilter_textureness(const DevMem2D& input, int winsz, float avgTexturenessThreshold, const DevMem2D& disp)
extern "C" void postfilter_textureness(const DevMem2D& input, int winsz, float avgTexturenessThreshold, const DevMem2D& disp, const cudaStream_t & stream)
{
avgTexturenessThreshold *= winsz * winsz;
@@ -551,8 +559,15 @@ namespace cv { namespace gpu { namespace bm
size_t smem_size = (threads.x + threads.x + (winsz/2) * 2 ) * sizeof(float);
stereobm_gpu::textureness_kernel<<<grid, threads, smem_size>>>(disp.ptr, disp.step, winsz, avgTexturenessThreshold, disp.cols, disp.rows);
cudaSafeCall( cudaThreadSynchronize() );
if (stream == 0)
{
stereobm_gpu::textureness_kernel<<<grid, threads, smem_size>>>(disp.ptr, disp.step, winsz, avgTexturenessThreshold, disp.cols, disp.rows);
cudaSafeCall( cudaThreadSynchronize() );
}
else
{
stereobm_gpu::textureness_kernel<<<grid, threads, smem_size, stream>>>(disp.ptr, disp.step, winsz, avgTexturenessThreshold, disp.cols, disp.rows);
}
cudaSafeCall( cudaUnbindTexture (stereobm_gpu::texForTF) );
}

10
modules/gpu/src/stereobm_gpu.cpp
View File

@@ -62,8 +62,8 @@ namespace cv { namespace gpu
{
//extern "C" void stereoBM_GPU(const DevMem2D& left, const DevMem2D& right, const DevMem2D& disp, int ndisp, int winsz, const DevMem2D_<uint>& minSSD_buf);
extern "C" void stereoBM_GPU(const DevMem2D& left, const DevMem2D& right, const DevMem2D& disp, int ndisp, int winsz, const DevMem2D_<uint>& minSSD_buf, const cudaStream_t & stream);
extern "C" void prefilter_xsobel(const DevMem2D& input, const DevMem2D& output, int prefilterCap = 31);
extern "C" void postfilter_textureness(const DevMem2D& input, int winsz, float avergeTexThreshold, const DevMem2D& disp);
extern "C" void prefilter_xsobel(const DevMem2D& input, const DevMem2D& output, int prefilterCap /*= 31*/, const cudaStream_t & stream);
extern "C" void postfilter_textureness(const DevMem2D& input, int winsz, float avgTexturenessThreshold, const DevMem2D& disp, const cudaStream_t & stream);
}
}}
@@ -115,8 +115,8 @@ static void stereo_bm_gpu_operator ( GpuMat& minSSD, GpuMat& leBuf, GpuMat& ri
leBuf.create( left.size(), left.type());
riBuf.create(right.size(), right.type());
bm::prefilter_xsobel( left, leBuf);
bm::prefilter_xsobel(right, riBuf);
bm::prefilter_xsobel( left, leBuf, 31, stream);
bm::prefilter_xsobel(right, riBuf, 31, stream);
le_for_bm = leBuf;
ri_for_bm = riBuf;
@@ -125,7 +125,7 @@ static void stereo_bm_gpu_operator ( GpuMat& minSSD, GpuMat& leBuf, GpuMat& ri
bm::stereoBM_GPU(le_for_bm, ri_for_bm, disparity, ndisp, winSize, minSSD, stream);
if (avergeTexThreshold)
bm::postfilter_textureness(le_for_bm, winSize, avergeTexThreshold, disparity);
bm::postfilter_textureness(le_for_bm, winSize, avergeTexThreshold, disparity, stream);
}